For Telecom operators, Customer churn, i.e., the event when the Customers leave a service provider, becomes a critical concern, studies have shown that acquiring new Customers cost five times more than to retain them. In competitive markets, where is increasingly important, to sustain growth as well as profitability correctly predicting the tendencies for Customer churn is important. Traditional predictive fashions frequently underperform due to the complex nature of client behavior. In this examine, we introduce a unique composite deep mastering framework whose hyperparameters are optimized the usage of the Particle Swarm Optimization (PSO) set of rules. Our method integrates a couple of neural community architectures to effectively capture each spatial and temporal patterns in client interactions. The PSO set of rules systematically first-rate-tunes parameters including activation functions, regularization techniques, gaining knowledge of rates, optimizers, and neuron counts—ensuing in a model that demonstrates robust overall performance. We evaluated our approach the usage of key metrics consisting of accuracy, precision, recollect, F1 score, and ROC AUC on a numerous purchaser dataset. Comparative analyses were conducted in opposition to established deep studying fashions (LSRM_GRU, LSTM, GRU, CNN_LSTM) in addition to other conventional methods (KNN, XG_BOOST, DEEP BP-ANN, BiLSTM-CNN, and Decision Tree). Experimental results stompy that our PSO-enhanced composite deep learning model stands out significantly compared with conventional models. Comparing the ROC-AUC scores of 0.932 and 0.93, F1 scores of 0.90 and 0.895, and accuracy rates of 83.2% and 93% on both Cell2Cell and IBM Telco datasets. it is indeed effective for practical churn prediction use incitements efficiencies. Var The experimental results demonstrate that our PSO express tree model outperforms conventional methods, achieving better performance with ROC totter score above 0.932 and 0.93, F 1 scores above 0.90 and 0.895 as well as accuracy rates in excess of 83.2% (% paper) and 93% (on the Telco data set) for Cell2Cell and IBM Telco respectively. This is further confirmation of its effectiveness and promise for practical churn prediction applications.